A method for quantizing a machine learning model during an inference phase, including determining a normalization factor using a set of floating-point values and a damped value of a damped value sequence; and assigning a quantized value for each floating-point value of the set of floating-point values based on the damped value sequence and the normalization factor.
Legal claims defining the scope of protection. Each claim is shown in both the original legal language and a plain English translation.
2. The method of claim 1, further comprising determining the damped value from the damped value sequence using the damped value sequence and a number of quantization bits.
A method for processing signals involves generating a damped value sequence from an input signal, where the sequence is derived by iteratively applying a damping factor to the signal. The method further includes determining a specific damped value from this sequence by considering both the sequence itself and a predefined number of quantization bits. The quantization bits define the resolution or precision used to represent the damped values, ensuring that the selected value is appropriately quantized for further processing or output. This approach is useful in applications where signal processing requires controlled attenuation or filtering, such as in audio processing, communication systems, or sensor data analysis. The method ensures that the damped value is accurately represented within the constraints of the available quantization bits, maintaining signal integrity while reducing computational complexity. The iterative damping process allows for fine-grained control over signal attenuation, while the quantization step ensures compatibility with digital processing systems. This technique is particularly valuable in systems where precise signal manipulation is required under resource constraints.
3. The method of claim 2, wherein determining the damped value from the damped value sequence comprises using the largest damped value from the damped value sequence based on the number of quantization bits.
6. The method of claim 1, wherein the normalization factor is determined using a maximum value from the set of floating-point values, a minimum value from the set of floating-point values, and the damped value.
11. The computer-readable medium according to claim 10, wherein determining the damped value from the damped value sequence comprises using the largest damped value from the damped value sequence based on the number of quantization bits.
14. The computer-readable medium according to claim 9, wherein the normalization factor is determined using a maximum value from the set of floating-point values, a minimum value from the set of floating-point values, and the damped value.
18. The quantizer of claim 17, wherein the damped value from the damped value sequence is determined using the damped value sequence and a number of quantization bits.
This invention relates to signal processing, specifically to a quantizer used in digital signal processing systems. The problem addressed is the need for efficient quantization of signals while minimizing distortion and preserving signal quality, particularly in applications where computational resources are limited. The quantizer processes an input signal by generating a damped value sequence, which is a series of values that gradually reduce the amplitude of the input signal over time. This sequence is used to determine a final damped value, which is then quantized into a digital representation using a specified number of quantization bits. The number of quantization bits controls the resolution of the output, allowing for a trade-off between precision and computational efficiency. The damped value sequence is derived from the input signal, and the final damped value is selected based on this sequence and the quantization bit depth. This approach ensures that the quantization process adapts dynamically to the signal characteristics, reducing distortion while maintaining computational efficiency. The method is particularly useful in audio processing, communication systems, and other applications where signal fidelity and resource constraints are critical. The invention provides a flexible and efficient way to quantize signals while balancing performance and resource usage.
Cooperative Patent Classification codes for this invention. Click any code to explore related patents in that topic.
November 7, 2018
October 18, 2022
Browse 5M+ US patents with plain-English claim translations and AI-generated analysis.